Study: Machine learning model improves perioperative pain assessment

Editor's Note

A new machine learning model using photoplethysmogram (PPG) data more accurately assesses pain during and after surgery compared to existing commercial methods, according to research published January 24 in Nature’s npj Digital Medicine.

Analyzing data from 242 patients, researchers developed an XGBoost-based model to assess intraoperative and postoperative pain and compared it to the widely used Surgical Pleth Index (SPI). The new model demonstrated strong performance, with an area under the receiver operating characteristic curve (AUROC) of 0.819 for intraoperative pain and 0.927 for postoperative pain, compared to SPI’s 0.829 and 0.577, respectively. The findings suggest that machine learning can significantly enhance perioperative pain monitoring, researchers write, particularly in postoperative settings where SPI struggles to differentiate between pain states.

These findings are a result of collecting PPG data at six time points: preoperatively, before and after intubation, before and after skin incision, and postoperatively. Researchers assessed pain using numerical rating scales or hemodynamic responses, then extracted 59 waveform features from the PPG signal to identify key indicators of pain for each surgical phase. The combined perioperative model incorporated the top features from both intraoperative and postoperative assessments to create a unified pain prediction system.

Study authors caution that labeling remains a challenge due to the subjective nature of pain perception, and intraoperative pain was inferred based on hemodynamic changes rather than direct patient feedback. The study also did not compare the new model to other commercial indices, which may have provided additional context. Additionally, external validation across diverse surgical populations is needed to assess generalizability. Nonetheless, these findings demonstrate the potential of machine learning to improve perioperative pain assessment by integrating multiple physiological indicators and adapting across surgical phases.